Automatic Feature Extraction to COSMOS-7 Content Models

Abstract

Extracting and modeling multimedia content in MPEG-7 is a laborious task as most systems use manual or partially automatic extraction methods. Systems that use partially automatic extraction techniques tend to use a limited set of features that are of little or no use to other applications negating the purpose of the MPEG-7 standard. In this paper we present our automatic extraction methods for the structural components of media. In particular, we extract from scenes, shots, objects and spatial and temporal relationships. These features are then processed and incorporated into an MPEG-7 content model that is rich enough to facilitate content filtering.